In this paper, sensor data are collected by group wise perception for preprocessing and fusion, SSA method is introduced to divide the output of LSTM into temperature loading effect with periodic trend, and residual data as vehicle loading effect and noise, combined with BDLM method to reduce the noise, increase the accuracy and stability of the model, and effectively monitor the bridge structure in real time. Then the Stacking integrated learning algorithm is used to mix different kinds of base learners to reduce the variance, which effectively improves the generalization ability of the model and realizes the fault warning of the bridge structure. The results show that the proposed method can effectively reduce noise, increase the accuracy and stability of the model, and alleviate the risk of overfitting. The LSTM-SSA-BDLM model can obtain vehicle-induced strain data under lossy and nondestructive conditions, and the four damage assessment indexes of “k, R², b and Ta” are stably distributed in the range of 0.45~1.55, which can effectively identify the hypothetical bridge damage. The baseline value of the warning threshold is pre-tested and estimated using the Pareto distribution model, and the value of the mid-span disturbance for a suspension bridge with 95% guarantee is obtained as -0.7076m, which ensures that the baseline value of the threshold can meet the standard of material strength.
